Skip to content

Latest commit

 

History

3 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

llm-cost-benchmark

Public, reproducible LLM cost benchmarks from Harpd. The headline metric is cost per successful task — not per-call price, not per-token cost — because that is the number production systems actually pay against.

First benchmark: JSON extraction (100 tasks)

Structured JSON extraction from messy natural-language input (English + Japanese + Chinese). 8 models, 100 real tasks, temperature 0.

Model Success $ / task $ / successful task
gpt-4o-mini 88% $0.00096 $0.001091
deepseek-v3 91% $0.00174 $0.001912
gemini-2.5-flash 90% $0.00270 $0.003000
llama-3.1-70b 84% $0.00405 $0.004819
claude-haiku-4 93% $0.00560 $0.006022
gemini-2.5-pro 95% $0.01100 $0.011579
gpt-4.1 96% $0.01280 $0.013333
claude-sonnet-4 97% $0.02100 $0.021649

The numbers above are a methodology preview (modeled from public 2026 list prices + assumed success/latency). Replace them with live runs:

cp runner/.env.example runner/.env   # add OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, DEEPSEEK_API_KEY, TOGETHER_API_KEY
node runner/index.mjs

The runner writes fresh data/*.json and data/*.csv. Everything is reproducible on a clean machine.

Repository layout

data/          benchmark results (JSON + CSV, one file per run)
pricing/       list-price snapshot the benchmark cost math used
runner/        open-source benchmark runner (live runs overwrite data/)
scripts/       analysis helpers
DATA_DICTIONARY.md   field-level definitions for every dataset
METHODOLOGY.md       collection, validation, execution, scoring, limitations, update frequency, reproduction

Website ↔ data

This repository and harpd.com mirror each other:

The GitHub repo is the source of truth for raw benchmark artifacts and the runner; harpd.com serves the live machine-readable exports and the human-readable summaries. Every last-updated date on the site comes from the data itself (generatedAt / verified_at), never from a build.

Why we publish the data

The cheapest model that succeeds on your real tasks is the only model that matters. Publishing raw CSV/JSON is the only way the community can verify that claim — and the only way Harpd ModelSwitch can earn the authority it needs.

License

MIT.


About Harpd

Harpd is the AI Cost Intelligence platform for the agent era — measure, optimize and control production AI spend, from cost per successful task to agent-payment budgets (x402 / USDC).

About

Public, reproducible LLM cost benchmarks from Harpd (harpd.com) — ranked by cost per successful task, not per-call price.

Topics

Resources

Stars

4 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages